Learning Curve Dynamics with Artificial Neural Networks

Abstract

Our objective is to learn the natural curve shapes with the help of Artificial Neural Networks (ANN). Our research aims to improve curve dynamics generated by the parametric models and the PCA. By running the ANN on the dataset of historically observed term structures of forward commodity prices we learn how the curves evolve over time. This allows us to determine stable natural shapes of the curves as well as to detect statistically significant deviations from the expected, predicted behaviour.

Keywords: Artificial Neural Networks, Term Structure, Curve Dynamics

JEL Classification: C61, C63, G13, G15